I say real circuits do not work like ideal maths equations full stop. Whether the difference is significant or not is another matter. In my experience, you have to get an awful lot of details right to get acceptable insignificance.
I know this from experience. But before you take aim, I haven't blamed feedback in and of itself...however, the application of feedback in real circuits can and does excite unwanted side-effects and this requires very careful attention.
I wasn't criticizing Bruno for using NFB; not least because he probably had no choice. I think you may be inviting me to spar with you over semantics. If a circuit is "linear" then linear maths applies. In my experience, real circuits aren't linear.
This is the sweeping statement from the paper that I do not feel is proven by the analysis in the paper:
I have heard some of Bruno´s 5th order Ncore amps (Jeff Rowland uses them in their 70k$ monoblocks) and, and I say this with some trepidation, initially you have the impression that it sounds different to what you are used to. My own paX amp is extremely clean, yet Bruno´s sounded also very clean yet still different to me. But when I tried to identify the usual markers to characterise the sound I couldn´t find anything missing or superfluous.
I also spoke to the people that were subjected to his flight of fancy, a tube amp with 60dB nfb at 20kHz, and they swear it sounded the best tube amp they ever heard, but I have no direct exposure to it.
I think it is safe to say it does something, but not sure what...
Jan
PS This was all serious but uncontrolled listening so use at your own peril!
Last edited:
Thanks - your "ear witness" report is invaluable.
It's always tricky when in a new environment without a familiar amp to compare with at the same time. Could you elaborate on its ability to play "music"...that is to be engaging (for want of a better term)?
That does not say anything.....
Nordacoustics class d amps are also based on 5th order Ncore; some € 1500 for a stereo amp.
It just illustrates the sometimes ridiculous prices going on in HiFi.
That is true, but also there's many different kinds of nCore's. The Kii speaker has 6 of the 250W variety in each box, and 'only' costs 10k/pair.
I think the Rowlands are 2.5kW.
Jan
Yes, that's the one "bomb proof" haha. I ended up leaving the 33K in place with a 47K to ground. So, loop gain was pretty low. I had no means to measure anything I'm afraid but nothing awful happened, didn't appear to be any oscillation, zobel resistor didn't get hot, and it sounded better to me, I can't be objective about that (unable to measure)
I would say they exactly do; you must see some measurements results on analog circuitry designed purely by simulation math, and then poured into silicon, and that up to GHz frequencies. 1-5% matching to the design target are very common, but then we don't use free stock models, but proprietary behavioural models delivered with the silicon design kit.
Not considering the corner cases (like what is going to happen when the amplifier runs into rails, effectively killing the open loop gain, hence fundamentally modifying the loop gain) is a different story, it's a design problem and a limitation of our quasi-linear models, so it is not realistic to expect such cases to be covered by default. Considering these corner cases is possible, but requires a different set of math and modelling tools. Kind of: it is easy to identify with an arbitrary precision the oscillation criteria for a Wien bridge oscillator using linear analysis, but predicting the oscillation amplitude is a different kettle of fish and can't be accurately determined using the same tools and models.
Very interesting. I used to be involved in mixed-signal ASIC development in the previous century (that sounds terrible!) but I'm not up on the latest technology. Are you a designer? Simulation accuracy of circuits in silicon is very good by the sound of it. I suppose you still have some Monte Carlo to do.
I find there are more variables in the relatively unrefined environment of the discrete audio power amplifier. Simulation models are obviously much more sophisticated than Bruno's gain equations in this paper. Do you know to what extent Hypex (or other) circuits are fabricated in silicon?
It's actually a fun experiment to put a WB oscillator in SPICE with an ideal poly source that has a tiny amount of cubic non-linearity. Give it tiny hit of charge and watch it settle. Barney Oliver wrote a paper on why some HP oscillators would not settle fast because the amp was accidentally too good or something like that.
Models have gotten a *wee* bit better in the past 20 years (and I haven't been involved in even the most basic of mask design since ~2008-2009). 🙂
But several commercially available packages can do EM parasitics from a board layout and feed that back into one's SPICE models. Admittedly above my level (which is very very low), and a bit beyond the *average* DIYAudio user's pay grade. And subject to GIGO, especially if your component models are incomplete. (Useful though in terms of learning trends, which is where so many of our layout adages come from) But we certainly have plenty of people who are pro's.
There are more than enough stories about needing new masks because something got missed or was different than sims, though.
But several commercially available packages can do EM parasitics from a board layout and feed that back into one's SPICE models. Admittedly above my level (which is very very low), and a bit beyond the *average* DIYAudio user's pay grade. And subject to GIGO, especially if your component models are incomplete. (Useful though in terms of learning trends, which is where so many of our layout adages come from) But we certainly have plenty of people who are pro's.
There are more than enough stories about needing new masks because something got missed or was different than sims, though.
If adding NFB to a linear circuit degrades it, then obviously linearity is not the problem as NFB will not affect it. Must be output impedance etc.? Or is it that by "practical, linear" you actually mean 'slightly nonlinear'?traderbam said:
If a circuit is nearly linear then linear maths can be applied, but carefully. Most real circuits (certainly in audio) are sufficiently linear that the nonlinearity can be handled via what is essentially a perturbation technique. Hence NFB can work.
By linear maths I assume you are talking about closed form solutions of simple circuits. SPICE transient analysis handles non-linearity just fine. As for modeling one of my favorite examples was a simulation of driving a 400MHz amplifier to the rail and bringing it back down to 0. The bench and sim matched almost exactly and the behavior was very messy and complicated.
In post #179 I was using "linear" to distinguish from "switching". I'll stop doing that or clarify next time.
Are you saying that Bruno's mathematical equations prove that Butterworth = NDFL in a real, non-linear power amplifier? He has used an error injection method to show how a linear system rejects an error signal.
Yes, I am talking about simple models for simple circuits used under simple conditions. For example, a fairly conventional audio amp which is not clipping or slew-rate-limiting. Under those conditions NFB theory predicts certain outcomes and NFB delivers these in practice. NFB sceptics vary in the type of smoke they blow, some more sophisticated than others, but they all end up denying that NFB works because they don't like the resultant sound. That misses the point of NFB, which is not to create a sound they like but to help reproduce a sound which the musician liked.
You must have missed what started all this, the uA741. The very premises in the abstracts of the time were flawed. If you take any amplifier with very high DC gain and single pole compensated as a general purpose building block making a 20dB gain circuit with it requires feedback. The feedback is what makes it work not the cause of the problem.
Dobkin published some ideas to fix the slew problem in 1969 with op-amps of the day (the ones that brought out the necessary pins) and Roberge had this in his textbook/class-notes already in 1970 when he was my teacher. The audio community just was not listening.
- Status
- Not open for further replies.